llm-provider · diff
git:20260425.9a46fd4 to git:20260428.99f2785
183 added, 83 removed. Audit A to A.
What this update changes in the checks
No check changed: every fixed check gives the same answer before and after this update.
A tripped check is a label, not a verdict: read the matched text and the lines below. This week's updates that newly trip a check.
---
name: llm-provider
- description: Adds a new LLM provider implementing LLMProvider interface from src/llm/types.ts with call() and stream() methods. Integrates config in src/llm/config.ts, factory in src/llm/index.ts, and error handling. Use when adding new provider, 'add LLM backend', integrating external API. Do NOT use for modifying existing providers or debugging provider issues.
+ description: Adds a new LLM provider implementing LLMProvider interface with call() and stream() methods. Integrates with provider factory in src/llm/index.ts, config detection in src/llm/config.ts, and error handling via tracking and recovery. Use when adding a new model backend, integrating a third-party LLM API, or extending LLM platform support. Do NOT use for fixing bugs in existing providers, modifying existing provider behavior, or changing the LLMProvider interface.
---
# LLM Provider
## Critical
- - **Interface contract**: Provider MUST implement `LLMProvider` from `src/llm/types.ts` — both `call(options: LLMCallOptions): Promise<string>` and `stream(options: LLMStreamOptions, callbacks: LLMStreamCallbacks): Promise<void>` are required
- - **No authentication in code**: API keys/tokens come from env vars only (e.g., `process.env.PROVIDER_API_KEY`). Never hardcode secrets
- - **Validation before integration**: Provider must pass a unit test in `src/llm/__tests__/` before wiring into config/factory
- - **Error handling**: All errors must extend or match the patterns in `src/llm/types.ts` (e.g., `LLMError` or provider-specific exceptions)
+ 1. **All providers MUST implement the LLMProvider interface** from src/llm/types.ts with three methods:
+ - call(options: LLMCallOptions): Promise<string> — single non-streaming call returning text
+ - stream(options: LLMStreamOptions, callbacks: LLMStreamCallbacks): Promise<void> — streaming call invoking callbacks
+ - listModels?(): Promise<string[]> — optional; list available models from the API
- ## Instructions
+ 2. **Initialize client in constructor and store defaultModel from config**. Example: `this.client = new YourSDK({ apiKey: config.apiKey })`. Never lazy-initialize on first call — providers are instantiated once and cached in src/llm/index.ts.
- **Step 1: Implement the provider in src/llm/**
+ 3. **For EVERY response in call() and stream(), invoke trackUsage(model, usage)** from src/llm/usage.js before returning/ending. This is mandatory — it captures token metrics for CLI telemetry and cost analysis. If the API doesn't return usage data, estimate via estimateTokens(text), which assumes ~4 chars per token.
- Create `src/llm/{provider-name}.ts` implementing `LLMProvider<TConfig>` from `src/llm/types.ts`.
+ 4. **Both call() and stream() must respect the model parameter** using pattern: `options.model || this.defaultModel`. Never hardcode model names. Callers supply model overrides via LLMCallOptions.model.
- Reference existing providers:
- - `anthropic.ts` — streaming via `message_start`, `content_block_delta` events
- - `openai-compat.ts` — standard OpenAI-compatible API
- - `cursor-acp.ts` — headless agent mode with `--stream-partial-output`
+ 5. **Error handling: catch all errors, preserve error messages unchanged**. The retry logic in src/llm/index.ts handles transient errors (ECONNRESET, socket hang up, 529 overload). For seat-based providers (Cursor, Claude CLI), wrap stderr via parseSeatBasedError() for user-friendly messages.
- Implement:
+ 6. **Always update ProviderType union** (Step 2), DEFAULT_MODELS (Step 4), and createProvider() switch case (Step 5) in lock-step. Missing any one breaks the build or causes runtime Unknown provider error.
+
+ ## Instructions
+
+ ### Step 1: Create provider class file
+ Verify directory exists: `ls -la src/llm/`. Create `src/llm/your-provider.ts`. Match existing provider patterns (src/llm/anthropic.ts, src/llm/openai-compat.ts).
+
+ Minimal structure:
```typescript
- export class MyProvider implements LLMProvider<MyConfig> {
- constructor(config: MyConfig) { this.config = config; }
-
+ import type { LLMProvider, LLMCallOptions, LLMStreamOptions, LLMStreamCallbacks, LLMConfig, TokenUsage } from './types.js';
+ import { trackUsage } from './usage.js';
+ import { estimateTokens } from './utils.js';
+
+ export class YourProviderProvider implements LLMProvider {
+ private client: YourSDKType;
+ private defaultModel: string;
+
+ constructor(config: LLMConfig) {
+ if (!config.apiKey) throw new Error('API key required');
+ this.client = new YourSDK({ apiKey: config.apiKey, ...(config.baseUrl && { baseURL: config.baseUrl }) });
+ this.defaultModel = config.model;
+ }
+
async call(options: LLMCallOptions): Promise<string> {
- // Non-streaming call
- // Return full text response
+ const model = options.model || this.defaultModel;
+ const response = await this.client.messages.create({ model, max_tokens: options.maxTokens || 4096, system: options.system, messages: [{ role: 'user', content: options.prompt }] });
+ trackUsage(model, { inputTokens: response.usage?.input_tokens || 0, outputTokens: response.usage?.output_tokens || 0 });
+ return response.content?.[0]?.text || '';
}
async stream(options: LLMStreamOptions, callbacks: LLMStreamCallbacks): Promise<void> {
- // Streaming call — invoke callbacks.onText() for each chunk
- // Call callbacks.onEnd() when complete, callbacks.onError() on error
+ const model = options.model || this.defaultModel;
+ const messages = [...(options.messages || []), { role: 'user' as const, content: options.prompt }];
+ try {
+ const stream = await this.client.stream({ model, max_tokens: options.maxTokens || 10240, system: options.system, messages });
+ let stopReason: string | undefined, usage: TokenUsage | undefined;
+ for await (const chunk of stream) {
+ if (chunk.delta?.text) callbacks.onText(chunk.delta.text);
+ if (chunk.delta?.stop_reason) stopReason = chunk.delta.stop_reason;
+ if (chunk.usage) usage = { inputTokens: chunk.usage.input_tokens, outputTokens: chunk.usage.output_tokens };
+ }
+ if (usage) trackUsage(model, usage);
+ callbacks.onEnd({ stopReason, usage });
+ } catch (error) { callbacks.onError(error instanceof Error ? error : new Error(String(error))); }
}
}
```
- **Verify**: Type-check passes (`npx tsc --noEmit`) and provider exports correctly.
-
- **Step 2: Add configuration in src/llm/config.ts**
-
- Register provider type and config interface in `src/llm/config.ts` alongside existing providers (`anthropic`, `openai`, `cursor`, `claude-cli`).
+ Verify: File exports the class; imports match existing providers.
- Add to union type:
+ ### Step 2: Add to ProviderType union
+ Edit `src/llm/types.ts` line 1. Add your provider in kebab-case:
```typescript
- export type LLMConfig = AnthropicConfig | OpenAIConfig | MyProviderConfig | ...;
+ export type ProviderType = 'anthropic' | 'vertex' | 'openai' | 'cursor' | 'claude-cli' | 'your-provider';
```
+ Verify: `npx tsc --noEmit` shows no ProviderType errors.
- Add config interface:
+ ### Step 3: Add config fields
+ If your provider needs fields beyond apiKey, model, baseUrl, extend LLMConfig in src/llm/types.ts:
```typescript
- export interface MyProviderConfig {
- provider: 'my-provider';
- apiKey?: string; // optional if using env var
+ export interface LLMConfig {
+ provider: ProviderType;
model: string;
- // other settings
+ fastModel?: string;
+ apiKey?: string;
+ baseUrl?: string;
+ yourProviderSecret?: string;
}
```
- **Verify**: Config merges into `LLMConfig` union without conflicts.
+ ### Step 4: Update config.ts
+ Edit `src/llm/config.ts`:
- **Step 3: Wire into factory in src/llm/index.ts**
+ **Line 9:** Add to DEFAULT_MODELS:
+ ```typescript
+ export const DEFAULT_MODELS: Record<ProviderType, string> = {
+ anthropic: 'claude-sonnet-4-6',
+ vertex: 'claude-sonnet-4-6',
+ openai: 'gpt-5.4-mini',
+ cursor: 'sonnet-4.6',
+ 'claude-cli': 'default',
+ 'your-provider': 'your-provider/default-model',
+ };
+ ```
- Update `src/llm/index.ts` `createLLMClient()` function to instantiate provider:
+ **Line 17:** Add to MODEL_CONTEXT_WINDOWS if known:
+ ```typescript
+ export const MODEL_CONTEXT_WINDOWS: Record<string, number> = {
+ 'your-provider/model-name': 128_000,
+ };
+ ```
+ **Line 59:** In resolveFromEnv(), add env detection before final return null:
```typescript
- case 'my-provider':
- return new MyProvider(config as MyProviderConfig);
+ if (process.env.YOUR_PROVIDER_API_KEY) {
+ return {
+ provider: 'your-provider',
+ apiKey: process.env.YOUR_PROVIDER_API_KEY,
+ model: process.env.CALIBER_MODEL || DEFAULT_MODELS['your-provider'],
+ baseUrl: process.env.YOUR_PROVIDER_BASE_URL,
+ };
+ }
```
- Add import at top:
+ **Line 115:** In readConfigFile() validation, add 'your-provider' to includes list.
+
+ Verify: `npm run test -- src/llm/__tests__/ -t config` confirms env var detection works.
+
+ ### Step 5: Register in factory
+ Edit `src/llm/index.ts`. Add import (line ~4):
```typescript
- import { MyProvider } from './my-provider.js';
+ import { YourProviderProvider } from './your-provider.js';
```
- **Verify**: Factory passes config correctly and type-checks.
+ In createProvider() switch (line ~24), add before default case:
+ ```typescript
+ case 'your-provider':
+ return new YourProviderProvider(config);
+ ```
- **Step 4: Write unit test in src/llm/__tests__/**
+ Verify: `npx tsc --noEmit` passes; no type errors on switch cases.
- Create `src/llm/__tests__/my-provider.test.ts` with:
- - Mock API responses matching provider's protocol
- - `call()` test: verify request format, response parsing
- - `stream()` test: verify chunk parsing, event handling
- - Error handling test (invalid API key, timeout, malformed response)
+ ### Step 6: Write tests
+ Create `src/llm/__tests__/your-provider.test.ts`:
+ ```typescript
+ import { describe, it, expect, beforeEach } from 'vitest';
+ import { YourProviderProvider } from '../your-provider.js';
- Use existing test patterns from `anthropic.test.ts` or `openai-compat.test.ts`.
+ describe('YourProviderProvider', () => {
+ let provider: YourProviderProvider;
+ beforeEach(() => {
+ provider = new YourProviderProvider({ provider: 'your-provider', model: 'test', apiKey: 'test' });
+ });
- **Verify**: `npm run test -- src/llm/__tests__/my-provider.test.ts` passes.
+ it('implements LLMProvider interface', () => {
+ expect(typeof provider.call).toBe('function');
+ expect(typeof provider.stream).toBe('function');
+ });
- **Step 5: Test integration**
+ it('call() returns string', async () => {
+ const result = await provider.call({ system: 'helpful', prompt: 'hi' });
+ expect(typeof result).toBe('string');
+ });
- Run full LLM test suite:
- ```bash
- npm run test -- src/llm/
+ it('stream() invokes callbacks', async () => {
+ const texts: string[] = [];
+ let ended = false;
+ await provider.stream({ system: 'helpful', prompt: 'hi' }, {
+ onText: (t) => texts.push(t),
+ onEnd: () => { ended = true; },
+ onError: () => {},
+ });
+ expect(ended).toBe(true);
+ });
+ });
```
- Test config loading:
+ Verify: `npm run test -- src/llm/__tests__/your-provider.test.ts` passes.
+
+ ### Step 7: Integration test
+ Run factory tests with your provider env var:
```bash
- npm run build && npx caliber config --check
+ YOUR_PROVIDER_API_KEY=test npm run test -- src/llm/__tests__/index.test.ts
```
-
- If provider requires user setup (interactive prompts), add to `src/commands/interactive-provider-setup.ts` following patterns for Anthropic/OpenAI.
+ Verify: getProvider() instantiates your provider; llmCall() dispatches correctly.
## Examples
- **User says**: "Add support for Groq API"
+ ### Example 1: Local LM Studio server
+ User says: "I need caliber to use my local LM Studio instance."
- **Actions**:
- 1. Create `src/llm/groq.ts` → implement `LLMProvider` with OpenAI-compatible fetch calls
- 2. Update `src/llm/config.ts` → add `GroqConfig` type with `model`, `apiKey` fields
- 3. Update `src/llm/index.ts` → factory case for `'groq'`
- 4. Write `src/llm/__tests__/groq.test.ts` → mock Groq API responses, test `call()` and `stream()`
- 5. Run: `npm run test -- src/llm/groq.test.ts` → passes
+ Actions: Create src/llm/lm-studio.ts extending OpenAICompatProvider. Add 'lm-studio' to ProviderType. In config.ts:
+ ```typescript
+ if (process.env.LM_STUDIO_BASE_URL) {
+ return { provider: 'lm-studio', apiKey: '', model: 'local', baseUrl: process.env.LM_STUDIO_BASE_URL };
+ }
+ ```
+ Register in createProvider() case. User: `export LM_STUDIO_BASE_URL=http://localhost:8000/v1`. Result: caliber uses local LM Studio; tokens estimated via estimateTokens().
- **Result**: User can set `LLM_PROVIDER=groq` and `GROQ_API_KEY=...`, caliber uses Groq for all LLM calls.
+ ### Example 2: Ollama (seat-based)
+ User says: "Ollama is auto-detected; no API key needed."
+ Actions: Create src/llm/ollama.ts extending OpenAICompatProvider. Add 'ollama' to ProviderType and SEAT_BASED_PROVIDERS. In config.ts:
+ ```typescript
+ if (process.env.OLLAMA_HOST) {
+ return { provider: 'ollama', model: 'mistral', baseUrl: process.env.OLLAMA_HOST || 'http://localhost:11434/v1' };
+ }
+ ```
+ Result: Offline per-machine LLM without API keys.
+
## Common Issues
- **"Cannot find module './my-provider'"**
- - Verify file is named `src/llm/{provider-name}.ts` (kebab-case)
- - Import uses `.js` extension: `import { MyProvider } from './my-provider.js'`
- - Run `npm run build` to check TypeScript
+ **Unknown provider: your-provider**
+ - Cause: ProviderType updated but createProvider() case missing.
+ - Fix: Add case in src/llm/index.ts and import the class.
- **"MyProvider does not implement LLMProvider"**
- - Ensure both `call()` and `stream()` methods exist and match signature in `src/llm/types.ts`
- - `call(options: LLMCallOptions)` returns `Promise<string>`; `stream(options: LLMStreamOptions, callbacks: LLMStreamCallbacks)` returns `Promise<void>`
- - Run `npx tsc --noEmit` for full type errors
+ **Cannot find module './your-provider.js'**
+ - Cause: File named your_provider.ts (underscore) not your-provider.ts (kebab).
+ - Fix: Rename file to use kebab-case.
- **"Provider not found in factory"**
- - Check `src/llm/index.ts` `createLLMClient()` has `case 'my-provider'` matching config type
- - Verify config union in `src/llm/config.ts` includes `MyProviderConfig`
+ **API key is required for YourProvider**
+ - Cause: Env var YOUR_PROVIDER_API_KEY not set; resolveFromEnv() didn't detect it.
+ - Fix: Verify env var name in config.ts matches. Test: `YOUR_PROVIDER_API_KEY=test npm run test -- src/llm/__tests__/index.test.ts`.
- **"API key undefined at runtime"**
- - Provider reads from `process.env.PROVIDER_API_KEY` or `config.apiKey`
- - User must set env var or pass in config
- - Check `.env` file or shell: `echo $PROVIDER_API_KEY`
+ **LLM response did not include usage tokens**
+ - Cause: Provider API doesn't return usage (local models).
+ - Fix: Estimate tokens: `trackUsage(model, { inputTokens: estimateTokens(options.system + options.prompt), outputTokens: estimateTokens(response.text) });`
- **"Stream callbacks never fire; onEnd not called"**
- - Verify `stream()` calls `callbacks.onText()` for each chunk and `callbacks.onEnd()` when done
- - Ensure the async iterator is fully consumed before `callbacks.onEnd()` is called
- - Compare streaming pattern to `anthropic.ts` as reference
+ **Stream callbacks never fire; onEnd not called**
+ - Cause: Async iterator not fully consumed before method returns.
+ - Fix: Ensure iteration completes before onEnd(): `for await (const chunk of stream) { } callbacks.onEnd({ stopReason, usage });`
+
+ **My model parameter is ignored**
+ - Cause: call()/stream() doesn't use `options.model || this.defaultModel`.
+ - Fix: Replace hardcoded model: `const model = options.model || this.defaultModel; const response = await this.client.create({ model, ... });`
+
+ **trackUsage() is never called**
+ - Cause: Forgot to call trackUsage() in call() or stream().
+ - Fix: Add after every response: `trackUsage(model, { inputTokens: ..., outputTokens: ... });`
+
+ **Type error: Provider doesn't implement LLMProvider**
+ - Cause: Missing method or wrong signature.
+ - Fix: Verify all required methods exist with exact signatures from src/llm/types.ts. Copy-paste from anthropic.ts as reference.